How to Transcribe a Video Without a Link

There is no link field

This is not a hidden feature or a setting. The upload page carries one input, and its accept attribute is audio/*,video/*. There is no URL input anywhere in the page, and the endpoint behind it does not read one. We sent an Instagram Reel address at it three different ways to be sure:

• As a form field named url: 400 in 0.09 seconds, with the message Please select a non-empty audio or video file
• As a JSON body with the URL inside: 500, with Transcription failed. Please try another supported file
• A plain GET to the endpoint: 404 Not found
• The same clip as a file, 5,461,607 bytes: 200 in 5.85 seconds, 322 characters, 59 words

The first two come back fast because nothing was fetched. The request never went to Instagram, never looked at the address, and never queued anything. It read the part of the request where a file should have been, found no file, and stopped.

Why no tool can fetch it for you

People assume the tool is choosing not to fetch. It is not a choice. We requested https://www.instagram.com/reel/ today and got a 200 carrying two headers that close the door: X-Frame-Options: DENY and cross-origin-resource-policy: same-origin. The first keeps the page out of a frame. The second stops a script on another site from reading the media at all. A browser tool sitting on our domain cannot pull that video down, and neither can our server, which is why there is no fetch step in our code to find.

What to send instead

Send the file you already have, or make one. Three routes cover almost everything:

• Save or download the Reel, if the account allows it - you get a video file with the audio track inside
• Screen record it on your phone - you get a bigger file, and the audio is whatever your device captured
• Ask whoever posted it for the original - the cleanest audio you will ever get, and the fastest upload
• For a video you own already, skip all three and upload it as it is

Whatever route you take, the name has to end in one of nine endings: MP3, WAV, M4A, AAC, OGG, FLAC, MP4, MOV or WEBM. Anything else gets refused in under a tenth of a second, and nothing is uploaded when it does. If your phone produced an .opus, a .3gp or an .amr file, rename it to something on that list, or convert it - renaming works when the audio inside is already fine, which it usually is.

Cut the picture off first if the file is big

A Reel file is mostly picture. In the 20-second clip we used for these tests, the file is 16,318,601 bytes and the audio stream inside it is 296,876 bytes, which is 1.82%. The whole file uploads fine when it is small enough - our 5,461,607-byte test came back in 5.85 seconds - but a full-length Reel is tens of megabytes, and that is where uploads get slow and start failing. One command fixes it:

• ffmpeg -i yourfile.mp4 -vn -c:a copy audio.m4a - pulls the audio out untouched, under a second on our clips
• Upload audio.m4a instead - same words, a fraction of the bytes

How big is too big

Across the uploads we ran today, every file at or under 6.6 MB came back 200 on the first attempt, in 1.53 to 6.57 seconds. The 16.3 MB one is where it broke: 503 in 5.63 seconds on the first try, then 200 in 27.31 seconds on the retry, same file. There is no published limit we can point you at, so treat size as the thing that makes uploads slow and flaky rather than a wall you hit at an exact number. The audio out of that same 16.3 MB file is 295,093 bytes, and it came back in 3.10 seconds.

What comes back

The response has six fields: success, transcript, language, engine, fileSize and fileName. The text itself comes back as one block, with no timestamps, no speaker labels and no line breaks. There is no file waiting for you afterwards either - a GET to the same address returns 404 Not found - so hit TXT, SRT or VTT and save it before you close the tab. Our 20-second clip came to 59 words. Count yours against the length of the clip when you read it: about three words a second is what we measured on clean speech, and a transcript noticeably longer than that usually has a repeated block in it.